About :: TODO :: Blog :: RSS :: Old blog :: Projects :: GIT :: Gallery :: Notes

Sat, 28 Apr 2007

Great filesystem contest. Dbench talk.


Let's descuss a bit dbench benchmark, it is aimed to determine throughput of the file server accessed by several clients. It is mostly write operations, but there are directory and file reads too (for more details check client.txt file in the distro). Size of each operation varies from several kilobytes to several thousands of kilobytes. This is quite small size compared to another benchmark I ran, so it should obey the same observations, and generally it is true, except the fact that some major filesystems behave extremely poor in such workload, which was not observed in another tests. Tuned/default mount options have upto 30% difference for the winner filesystem, which in turn is upto 100% faster than the next one. Since it is mostly writes you might expect who is a winner...

Stay tuned, I did not yet ran reiser4 tests (hope I will be able to patch 2.6.20-rc tree with reiser4 patches from -mm tree) and did not setup described previously maildir benchmark. More tomorrow.

/devel/fs :: Link / Comments ()